On 20 Feb 2015, at 20:15, Viktor Dukhovni <ietf-dane@dukhovni.org> wrote:

On Fri, Feb 20, 2015 at 08:01:09PM +0100, Andreas Fink wrote:


Depends on what one  one means by "fallback". 

How about support (as a fallback) for older clients? How "safe" (no pun
intended) is it to disable as of today?

Its simple:  fallback = a MITM attacker can force fallback = youre pwned...

Fall back in the sense of we tell the end user it was transported in a secure way but actually insecure encryption was used by the server and the MITM was able to decode it. If its acceptable to transfer in the clear in case TLS fails is something the mail operator might choose or not and it depends also on the specific link. For example I could say I know google mail supports TLS so only TLS would be permitted in my config but then if TLS steps down to RC4, i'm no longer protected from that assumption.

 When RC4 is enabled
at a low preference MITM attackers cannot re-order the handshake

without invalidating the TLS "finished" message.

Why this? a MITM attack implies the man in the middle terminates the TLS and thus he would in above example "mimic" the google mailserver's behaviour and simply would only offer RC4 inbound. Your sending mailserver would accept that as only option and use RC4 to deliver. So there's no "reordering" as its the only option provided. On the outbound connection, he simply would use a secure connection as usual to pass through the commands.

I should be noted that, occasional bilateral security arrangements
aside, MTA to MTA SMTP is generally vulnerable to MiTM attacks
regardless of whether RC4 is enabled or not.

With DANE, SMTP client MTAs can also authenticate servers for which
no prior security settings exist, and in *that* case we have a
fairly MiTM resistant protocol.

In Postfix for peers that publish TLSA RRs, the "mandatory" TLS
protocol, cipher and exclusion lists apply.

By all means, try:

   smtp_tls_mandatory_protocols = !SSLv2, !SSLv3
   smtp_tls_mandatory_exclude_ciphers = RC4

If there are any domains that publish TLSA records for an SMTP
server that is capable only of legacy crypto, both they and I will
be surprised.

-- 
Viktor.

--